Working Group 3 Two-way model-data interaction Chair person: Matthew England WORKING GROUP OVERVIEW The scope of Working Group 3 discussions included the use of chemical tracers to assess/improve ocean models and using ocean models (often conceptual) to understand the distribution of chemical tracers. Issues left to other working groups included biogeochemical tracer modelling (WG2) and the assimilation of chemical tracer data into ocean models (WG4). The tasks set for this Working Group included describing the achievements and present status of chemical tracer modelling, detailing limitations and uncertainties, and making specific recommendations for future work and action items. The WG discussions in these three areas are synthesised below. ACHIEVEMENTS IN TRACER MODELLING o Chemical tracers used to assess/improve ocean models. Geochemical tracers are now widely used to assess the simulated circulation in ocean models. Tracers that have been used in this context include tritium, chlorofluorocarbons, natural and bomb-produced radiocarbon; and, to a lesser extent, oxygen, silicate, phosphate, isotopes of organic and inorganic carbon compounds and dissolved noble gases (e.g., helium and argon). Table 1 [http: //www.maths.unsw.edu.au/metweb/matthew/table_1.ps] from England and Maier- Reimer [1999] includes a list of these tracers and their various applications and properties. Particular aspects of ocean models that have been assessed in this way include Deep Western Boundary Current (DWBC) flow rates [e.g., Redler, 1997; England and Holloway, 1998], thermocline ventilation [e.g., Jia, 1996, Jia and Richards, 1996, Dutay, 1998], water-mass formation in the Southern Ocean including AABW flow rates [e.g., England, 1995; Robitaille and Weaver, 1995; England and Hirst, 1997, Haine et al., 1998], and deep ocean overturning in the Pacific Ocean using radiocarbon [e.g., Toggweiler et al, 1989; England and Rahmstorf, 1999; Duffy et al., 1997]. More details?????? Geochemical tracers have also provided a direct constraint on the apparent Peclet number ..... ??????????? [Bill Jenkins to complete] Jia and Richards o Ongoing incremental model improvement The above mentioned studies in ocean model assessment have contributed to a number of improvements in certain aspects of ocean GCMs. This is because model assessment using chemical tracers can give relatively unambiguous tests of model performance with regards to certain parameterisations or ventilation processes. The results of chemical tracer assessment studies then prompt certain model refinements, and in this way tracer data sets facilitate the process of ocean model improvement that is a key goal of WOCE. Examples include improved representation of subgrid-scale mixing effects [e.g., England, 1995; Robitaille and Weaver, 1995; England and Hirst, 1997, Duffy et al. 1997] and bottom boundary layer dynamics [Beckmann and Doescher, 1997]. In addition, geochemical tracers have helped to reveal fundamental model limitations with respect to horizontal resolution [Redler, 1997; ?????, 19??], interior T-S restoring techniques [Toggweiler et al, 1989], and the spatial extent of open-ocean convection [England and Hirst, 1997]. Coarse resolution models chronically exaggerate the spatial scales of open ocean convection and deep currents, while underestimating deep flow rates and diffusing downslope flows with excessive lateral mixing. More details?????? o Conceptual models used to.... ?????????? [Scott Doney to complete] - Interpret tracer age, Musgrave, Doney, Pickart, Thiele and Sarmiento, Sarmiento, Rhein - Estimate bulk mixing rates [Roether et al. 198??], DWBC recirculations, de/entrainment, - Estimate anthropogenic CO2 uptake (box-diffusion models) o PWP models of upper ocean ventilation and gas uptake ?????????? [Bill Jenkins to complete] Haine & Richards (1995) UNCERTAINTIES AND LIMITATIONS There are a number of limitations and uncertainties in the process of using chemical tracers to assess ocean models, and models to interpret observed tracer distributions. These are summarised here. o Input functions and boundary conditions There are a number of key areas of uncertainty in terms of forcing chemical tracer uptake in ocean models. These include the following: - Uncertainties in the parameterisation of the air-sea gas piston velocity k [Wanninkhof, 1992]. Estimates of the dependence of k on wind speed vary by up to a factor of two for typical open ocean wind speeds. This can result in significant uncertainties in the estimation of model air-sea gas exchange, particularly over deep convective mixed layers. This is critical as the chemical tracer signature in the deep ocean is directly tied to the simulated saturation levels in the winter mixed layer. This is more important for the slow equilibration gases such as CO2. - Uncertainties in the atmospheric histories of bomb-14C, CFCs, and CO2. Normally models assume some time history of tropospheric concentrations of bomb-14C, CFCs, and CO2, with a simple latitudinal dependence. The error bars on this assumption are probably relatively small. However, they are yet to be formally quantified. - Uncertainties in the input functions of tritium [Doney et al., 1993; Roether, 19??] and mantle helium [Farley et al., 1995]. There remains some source of uncertainty in the various components of tritium input into the World Ocean. Global estimates of tritium input should include some estimate of possible uncertainties due to unknowns such as precipitation rates, marine vapour exchange and river run-off. Mantle helium sources are very difficult to quantify directly. Moreover, the utility of mantle helium in ocean model assessment remains fundamentally limited by the degree of uncertainty surrounding the mantle helium flux from seafloor volcanism [e.g. Farley et al., 1995]. - Unknown error distributions in global climatologies of wind speed and sea-ice coverage (for forcing air-sea gas exchange). Air-sea gas fluxes of CFCs, 14C and CO2 depend on a knowledge of "climatological" wind speed and sea-ice. Yet these properties are poorly measured in some ocean regions, particularly in the Southern Ocean. Recent advances in remote sensing technology should improve these climatologies in the near future, although even then low-frequency climate variability might skew these estimates of global wind speed and sea-ice coverage. o Sampling errors and aliasing of temporal/spatial variability WOCE and non-WOCE tracer fields contain unknown sampling error distributions - both spatial and temporal. To directly quantify model skill, for example, to reject a model solution because it is inconsistent with tracer data, requires some knowledge of the sampling error distribution associated with that data [Haine and Gray, 1999]. This includes measurement error, which is perhaps often small, as well as errors associated with the aliasing of sub-grid scale spatial and temporal variability. Oceanic variability is ubiquitous and evident at many scales, both in space and time, so we can expect a degree of aliasing in the WOCE tracer data sets compared to what might be simulated in an ocean model. o Assessing ocean models in climate studies Ocean-only model assessment adopts climatological wind speed and sea-ice data to estimate air-sea gas fluxes. In coupled model studies, it is unclear whether model assessment should employ similar surface field climatologies or whether the internal model predicted winds and sea-ice are more appropriate. If the model fields are used, errors in the model generated sea-ice or winds could compensate for errors in the ocean circulation rendering the model validation inappropriate [as detailed by England and Maier-Reimer, 1999]. On the other hand, in some instances, such as over a spurious polynya in polar waters, using observed sea-ice fields could alias errors in the model predicted ocean circulation. It seems that both techniques should be adopted to assess ocean models within climate prediction systems. o Quantification of model skill, formulation of cost function. A range of different cost function definitions is required to identify the quantities that are most strongly constrained by tracer observations. These should include the best spatial and temporal fit to measurements and the best fit to model benchmarks (discussed below). Studies that seek to estimate quantities such as subduction rates, diffusivities and biogeochemical rates using inverse methods and tracer data are likely to be particularly fruitful (see also report to WG4). o Accelerated integration techniques in higher resolution models The cost of including additional passive tracers in coarse resolution ocean models is not overly expensive, even when multiple tracers are included. However, higher resolution models can be difficult to assess using chemical tracers when computational limits are already being pushed just in integrating with respect to T-S and circulation. It remains unclear to what extent the tracer integration can be optimised with regard to computational cost. Issues include: - Initialisation fields. Choosing an initial tracer solution that is close to the final equilibrated field [for tracers such as natural 14C] can minimise required computational costs [e.g., Aumont et al., 1998]. - Off-line tracer modelling techniques. These normally require careful consideration of internal model variability not normally resolved in the mean T-S and u-v fields saved for the off-line model. This includes consideration of model variability in convection fields and in eddy behaviour. Some error will be introduced into off-line models, and necessarily this is difficult to compute. o Regional models (transient tracer fields at open boundary conditions) Specifying boundary conditions for chemical tracers in regional ocean models can be problematic, especially for transient tracers. Even if a model domain is selected to have open boundaries that coincide with a WOCE hydrographic section, it is necessary to extrapolate the time- dependent transient tracer content at this open boundary, which requires assumptions about the long-term ocean circulation in particular regions. Examples of regional transient tracer modelling include that by Redler et al. [1998] and Barnier et al. [1997]. High resolution multi-tracer modeling is an effective tool to determine the seasonality and interannual variability in water mass production and transport. However, the high computational demands of such models presently restrict them to limited regions of the world ocean. Several high resolution model studies are underway, the French CLIPPER project [Barnier et al., 1997; CLIPPER, 1998] and the German FLAME program [Redler et al., 1998]. These projects use a variety of model configurations covering both the whole Atlantic and various smaller subdomains, with open boundary conditions to treat the in- and outflow of T-S and tracers. For these boundaries reliable tracer data are needed at the Drake Passage, 30E, 20S, and 70S to simulate the tracer signal in the model interior. For transient tracers such as CFCs, this requires an estimate of the temporal evolution of tracer at these sections from the 1930s to the present. The problem of regional transient tracer modelling is demonstrated in the North Atlantic model of Redler et al. [1998], wherein an open boundary exists at 18S. While the high resolution model is able to reproduce observed CFC patterns in the DWBC quite well, it fails to show up features in the equatorial current regime like the eastward flow of CFC enriched water at depths around 2000 m. In the model, advection of water with low CFC concentration from the south dilutes the near- equatorial waters, and this is entirely a boundary condition effect. Redler et al. [1998] did not have adequate estimates of CFC content at the southern open boundary to properly address this question. Similar problems show up in a model covering the whole Atlantic with open boundary conditions in the Drake Passage and along 30E (across the Agulhas Current). The CFC signal south of the Equator is diluted if the inflowing CFC signal from the open boundaries is missing. o Advection schemes: problems, requirements, solutions The centred-time centred-space (CTCS) scheme, traditionally used in GCMs, features conservation of mass and variance of the field combined with low computational costs. However, it has the side-effect of numerical dispersion and -- a main drawback for tracer-modelling -- it causes ripples and negative concentrations as the field approaches zero. Thus, it violates the second law of thermodynamics. The simple and cheap upstream scheme is not dispersive and it conserves mass, but at the cost of inducing high implicit diffusion, which makes it unsuitable for most model applications. Second-order schemes like QUICK and QUICKEST reduce the noise considerably, but unphysical extrema are only little affected, while the costs are about three times those of CTCS. To achieve sufficient accuracy for biogeochemical tracers, flux-correction schemes such as FCT [e.g., Gerdes et al., 1991], a combination of CTCS and upstream, are required. Iterative schemes based on upstream like MPDATA, which can preserve sign and produce smooth fields, would be ideal for tracers. To minimize the high costs of these schemes, "super-cycling" could be applied, i.e., a bigger timestep for passive tracers. o Some specific chemical tracer issues: - Radiocarbon: Normally modellers simulate radiocarbon as the deviation of the 14C/12C ratio from a standard atmospheric value. It is argued [Fiadiero, 1982] that this renders isotopic fractionation effects and biological conversion processes insignificant, so that radiocarbon is like a passive tracer which simply undergoes radioactive decay. Recent studies [Caldeira, 19?????] suggest that in fact .... ????? [Ken Caldiera to complete]. Another issue with radiocarbon is the separation of bomb-produced 14C from natural 14C in the observational data sets. Because radiocarbon was measured during GEOSECS as well as WOCE, some attempt should be made to quantify the time-dependent spreading of bomb-14C since the 1970s. - Argon-39: Being unaffected by biological processes and with a half-life of 269 years, argon-39 would be an ideal constraint on intermediate and deep ocean circulation, and a natural complement to the longer time-scale tracer 14C. However, only around 100 measurements of argon-39 have been made in the ocean, although most of these are in the Atlantic basin. Some modellers have used argon-39 as a constraint on ocean simulations [e.g., Orr, 19??], although the sparsity of measurements leaves this tracer more suited to simple model diagnosis, such as separating processes of water-mass conversion [e.g., Maier-Reimer, 1993]. o Uncertainties in derived data products Whilst modellers seek derived data products, such as integral quantities, column inventories, tracer fields objectively mapped onto isopycnal surfaces, 'age' estimates, and so on, the "gapiness" of tracer data (exacerbated when transient) renders many of these derived products subject to unknown error. Modellers need some measure of this error in order to use the derived data products judiciously. FUTURE WORK, ACTION ITEMS, RECOMMENDATIONS 1. Modellers need tracer field error distributions Estimates of the spatial and temporal variability of passive tracer fields are required for mapping profiles and sections and comparison of observations with model predictions. The sampling error introduced by surveys with measurements at WHP spacing is significantly greater than typical instrumental errors. In order to objectively estimate this sampling uncertainty the power spectrum of variability for each particular tracer is required. Current knowledge of these spectra are very poor and deserve attention in future. Preliminary work suggests that the spectra may depend only weakly on the tracer sources and sinks (Haine and Gray, 1999). If this is the case, there is a reasonable prospect that a universal passive tracer spectrum exists. This spectrum could be estimated directly on scales between 100 and 1000km using observations. There is also scope to estimate the power at shorter scales using eddy-resolving circulation models. Such studies are encouraged. 2. Tests of model "consistency" with observed tracer data Research that addresses whether, or not, passive tracer fields are consistent with tracer predictions from GCMs is preliminary to full-blown inverse studies using GCMs. The results from this work will identify the ways that tracer fields provide useful constraints on GCM circulation and mixing fields and will be helpful in guiding the future interpretation of tracer data. One theoretical way to approach this issue has been explored by Memery & Wunsch (1990; tritium) and Gray & Haine (1999, CFCs) in the North Atlantic. 3. Need improved input functions The use of chemical tracers in ocean models requires well-known input functions. Improved knowledge of tracer entry functions is required foremost for mantle helium, surface tritium, and radiocarbon. Improved estimation of the air-sea gas piston velocity is also needed for tracers such as CO2, CFCs and 14C. 4. Need gridded and derived tracer quantities for assessing model skill Ocean model assessment using chemical tracers relies on the availability of quality controlled hydrographic sections and data sets. The tracer modelling community endorses the continued processing of WOCE tracer data to this end. Much can be learnt from WOCE tracer section comparisons and water-mass analyses (e.g., property-property diagrams). Tracer observationalists and modellers should also work together to develop new integral measures of model fidelity. As discussed above, these should include error estimates for tests of model 'consistency'. An analogy in hydrographic data sets is the estimation of poleward heat transport to assess ocean models at WOCE transects. Recommended data products include integral quantities, column or water-mass inventories, and tracer fields objectively mapped onto isopycnal or geopotential surfaces. Where possible, normalised regional maps of transient tracer concentrations are also recommended. Modellers should liaise with tracer observationalists to target those products that are critical to capturing ocean climate processes. Where available, the use of time-differenced tracer fields (e.g., WOCE minus GEOSECS tritium and radiocarbon) is recommended in ocean model assessment. Such data products constrain ocean model solutions more than simple non-repeat hydrographic sections. 5. Need ongoing tracer measurements for assessing ocean models Whilst the WOCE tracer data sets can be combined with other non-WOCE tracer data to develop some picture of low-frequency variability in ocean circulation, this is substantially limited. In order to assess the decadal-scale ocean circulation in models it is strongly recommended that a strategy be developed for measuring chemical tracers in the global ocean at selected times subsequent to the WOCE tracer measuring programme. This should ideally be done in consultation with the ocean modelling community so that key oceanic processes relevant for climate models be measured. 6. Models should be used to - estimate the spatial-temporal covariance functions of tracers (related to recommendation 1 above). High resolution process-oriented models can be used to estimate the spatial and temporal scales of variability in tracer fields. This will improve our knowledge of tracer field sampling error distributions. - extrapolate-interpolate WOCE data, both in space and time. Adjoint methods could be useful in filling out the data gaps in WOCE tracer fields, for example to facilitate the reconstruction of open boundary conditions for regional ocean models or to estimate basin-scale inventories for use in ocean model assessment. - assess the validity of derived "age" data products. For example, modelled CFCs in combination with modelled age tracer fields can be used to test techniques for determining CFC-11/CFC-12 ages and to verify methods used to determine water masses and their relative contributions to the "age" value [e.g., England and Holloway, 1998]. - estimate meridional and air-sea fluxes of tracers. For example, model studies can be used to estimate air-sea fluxes of tracers over convective mixed layer regions where direct measurements are difficult to make. Also, global integral quantities such as the meridional transport of a tracer can be estimated in combination with hydrographic and modelling techniques. - examine tracer flow dynamics in process studies (e.g., convection, bottom boundary layer). Modelling studies that investigate the processes of passive tracer transport in oceanic boundary layers and the ocean interior are required to guide the interpretation of field measurements. Preliminary progress has been made on the mechanisms of CFC stirring and mixing in ocean mixed layers by Haine & Richards (1995) and Haine & Marshall (1998). Future work should address the role of interior mixing on tracers with qualitatively different sources and sinks including derived tracer quantities such as tracer age. 7. Recommend multiple tracers in ocean model assessment Because property-property analyses provide a more powerful assessment of ocean model skill, particularly when using properties that give relatively distinct information, multiple tracer modelling is recommended in ocean climate model assessment. A good example would be to simulate natural 14C, chlorofluorocarbons, tritium, and if a suitable input function is known, mantle helium. 8. Provide model output for tracer observationalists Interactions between the tracer observing and modelling communities are often one-way, with modellers seeking tracer data without necessarily returning model output results and other diagnostics to tracer observationalists. In order to enhance the two-way interaction between tracer modellers and those that collect tracer data, it is recommended that model output be accessible through an interactive Web site, perhaps housed from a WOCE IPO location. Modellers would be encouraged to submit their output to the site along with a brief READ_ME file that describes their model configuration. The Web facility would then enable the plotting of WOCE and non-WOCE sections, calculation of tracer inventories and other products, such as air-sea fluxes, tracers on isopycnal surfaces, and mixed layer saturation maps. REFERENCES Aumont, O., J.C. Orr, D. Jamous, P. Monfray, O. Marti, and G. Madec, A degradation approach to accelerate simulations to steady-state in a 3-D tracer transport model of the global ocean, Clim. Dyn., 14, 101-116. Barnier, B., Le Provost, C., Madec, G., Treguier, A.M., Molines, J.M., and Imbard, M.: The CLIPPER project: High resolution modelling of the ocean circulation in the Atlantic forced by or coupled with the atmosphere, In: Proceedings of the international symposium "Monitoring the oceans in the 2000s: an integrated approach." Biarritz, France, 15-17 Oct 1997, CNES/NASA. CLIPPER project: Modelisation a haute resolution de la circulation dans l'ocean Atlantique forcee et couplee ocean-atmosphere, Rapport Scientifique et Technique 1997, CLIPPER-R2-98, LEGI/LPO/LODYC, March 1998. Duffy, P.B., K. Caldeira, J.P. Selvaggi, and M.I. Hoffert, Effect of subgrid-scale mixing parameterizations on simulated distributions of natural 14C, temperature, and salinity in a three-dimensional ocean general circulation model, J. Phys. Oceanogr., 27, 498-523, 1997. Dutay, J-C., Influence du melange vertical et de la couche melangee sur la ventilation de l'ocean: simulations numeriques des traceurs transitoires tritium-helium-3 et CFC avec le modele OPA. PHD thesis, Universite Paris 6, 1998. England, M.H., Using chlorofluorocarbons to assess ocean climate models, Geophys. Res. Lett., 22, 3051-3054, 1995. England, M.H., and A.C. Hirst, Chlorofluorocarbon uptake in a World Ocean model, 2: Sensitivity to surface thermohaline forcing and subsurface mixing parameterization, J. Geophys. Res., 102, 15,709-15,731, 1997. England, M.H., and G. Holloway, Simulations of CFC-11 outflow and seawater age in the Deep North Atlantic, J. Geophys. Res., 103, 15,885-15,902, 1998. England, M.H., and S. Rahmstorf, Sensitivity of ventilation rates and radiocarbon uptake to subgrid-scale mixing in ocean models, J. Phys. Oceanogr., in press, 1999. England, M.H., and Maier-Reimer, E., Using chemical tracers in ocean models, Reviews of Geophysics, 1999, in press. Fiadiero, M.E., Three-dimensional modeling of tracers in the deep Pacific Ocean, 2, Radiocarbon and the circulation, J. Mar. Res., 40, 537-550, 1982. Farley, K.A., E. Maier-Reimer, P. Schlosser, and W.S. Broecker, Constraints on mantle 3He fluxes and deep-sea circulation from an ocean general circulation model, J. Geophys. Res., 100, 3829-3839, 1995. Gerdes, R., C. Kberle, and J. Willebrand, 1991: The influence of numerical advection schemes on the results of ocean general circulation models. Clim. Dynamics, 5, 211-226. Gray & Haine, (1999), Constraining an ocean general circulation model of the North Atlantic using chlorofluorocarbon observations. In prep. Haine, T.W.N., and K.J. Richards, The influence of the seasonal mixed layer on oceanic uptake of CFCs, J. Geophys. Res., 100, 10,727-10,744, 1995. Haine, T.W.N., A.J. Watson and M.I. Liddicoat and R.R. Dickson, The flow of Antarctic bottom water in the southwest Indian ocean estimated using CFCs, JGR, 103, 27637--27653, 1998. Haine, T.W.N., and J. Marshall, 1998, Gravitational, symmetric and baroclinic instability of the ocean mixed layer, J. Phys. Oceanogr., 28, 636-658. Haine & Gray, 1999, Quantifying meso-scale variability in ocean transient tracer fields. In prep. Jia, Y., 1996: On tracer derived ages in the Atlantic Isopycnic Model. International WOCE Newsletter, 23, 14-17. (Unpublished manuscript) Jia Y., and K.J. Richards, Tritium distributions in an isopycnic model of the North Atlantic, J. Geophys. Res., 101, 11883-11901, 1996. Maier-Reimer, E., Geochemical cycles in an ocean general circulation model; preindustrial tracer distributions, Glob. Biogeochem. Cycles, 7, 645-677, 1993. Memery & Wunsch, JGR, 95, 5239--5256, 1990. Redler, 1997: Prozesstudien zur Wassermassenbildung und -ausbreitung in numerischen Modellen des subpolaren Nordatlantik, PhD thesis, University of Kiel, pp 120. Redler,R., Klaus Ketelsen, Joachim Dengg & Claus W. Boening: A High-Resolution Numerical Model for the Circulation of the Atlantic Ocean, in: Proceedings of the Fourth European SGI/Cray MPP Workshop (Sep 10-11, 1998, IPP, Garching, Germany) Editors: Hermann Lederer and Friedrich Hertweck, p 95-108 Robitaille, D.Y., and A.J. Weaver, Validation of sub-grid scale mixing schemes using CFCs in a global ocean model, Geophys. Res. Lett., 22, 2917-2920, 1995. Toggweiler, J.R., K. Dixon, and K. Bryan, Simulations of radiocarbon in a coarse-resolution world ocean model. I: Steady state prebomb distributions. J. Geophys. Res., 94, 8217-8242, 1989. Wanninkhof, R., Relationship between wind speed and gas exchange over the ocean, J. Geophys. Res., 97, 7373-7382, 1992.